Topic/monitoring

[bosilca]

Add a new capability to Open MPI. With this patch, OMPI gain the capability to monitor all communications and to dump the communication heat map (including differences between internal and user communications) for point-to-point, collective and RMA. The entire framework is driven using the new, MPI Forum blessed, MPI_T interface.

Fully fledged documentation and scripts to manipulate the generated data are also provided.

[mike-dubman]

merge leftovers

[bosilca]

thanks. fixed.

[hjelmn]

:bot:retest:

[hppritcha]

lanl distcheck doesn't like this

make[3]: Entering directory `/home/hppritcha/jenkins/workspace/ompi_master_pr_distcheck/openmpi-gitclone/_build/ompi/mca/common/monitoring'
  CC       libmca_common_monitoring_la-common_monitoring.lo
  CC       libmca_common_monitoring_la-common_monitoring_coll.lo
  LN_S     libmca_common_monitoring.la
../../../../../ompi/mca/common/monitoring/common_monitoring.c:17:31: fatal error: common_monitoring.h: No such file or directory
 #include "common_monitoring.h"
                               ^
compilation terminated.
../../../../../ompi/mca/common/monitoring/common_monitoring_coll.c:17:31: fatal error: common_monitoring.h: No such file or directory
 #include "common_monitoring.h"
                               ^
compilation terminated.

[hppritcha]

botany bay doesn't like this either:

  CC       monitoring_prof.lo
monitoring_prof.c:385:6: warning: no previous prototype for function 'monitoring_prof_mpi_init_f2c' [-Wmissing-prototypes]
void monitoring_prof_mpi_init_f2c( MPI_Fint *ierr ) { 
     ^
monitoring_prof.c:394:6: warning: no previous prototype for function 'monitoring_prof_mpi_finalize_f2c' [-Wmissing-prototypes]
void monitoring_prof_mpi_finalize_f2c( MPI_Fint *ierr ) { 
     ^
In file included from monitoring_prof.c:417:
../../ompi/mpi/fortran/mpif-h/bindings.h:67:2: error: Unrecognized Fortran name mangling scheme
#error Unrecognized Fortran name mangling scheme
 ^
monitoring_prof.c:424:28: error: expected identifier
                           (MPI_Fint *ierr),
                           ^
monitoring_prof.c:427:1: error: expected function body after function declarator
OMPI_GENERATE_F77_BINDINGS (MPI_FINALIZE,
^
2 warnings and 3 errors generated.
make[2]: *** [monitoring_prof.lo] Error 1
make[1]: *** [install-recursive] Error 1
make: *** [install-recursive] Error 1
Build step 'Execute shell' marked build as failure

[hppritcha]

the sign off checker doesn't like a lot of these commits either.
Please squash down the commits in this PR at least some.

[bosilca]

I don't see how it cannot find the "common_monitoring.h" file. We are supposed to automatically include "." so the file should be readily available.

I am really sorry for the sign-off checker. The commits will be squashed anyway once merged.

[ggouaillardet]

@bosilca that was just a typo that causes common_monitoring.h is missing from the tarball
an other error is related to lack of weak symbols and no fortran bindings

these should be both fixes in #3131
once it passes CI, i will PR vs your branch so you can simply merge it

[bosilca]

Thanks @ggouaillardet, I haven't noticed the failure was on due "make dist".

[ggouaillardet]

@bosilca i filed bosilca#2 to fix all the glitches with this PR

[bosilca]

Thanks @ggouaillardet, I merged your PR.

[ggouaillardet]

@bosilca i filed bosilca#3 to silence some misc warnings